One of a series of six white papers for my client on the subject of AI and the different types of machine learning. Each acted as a companion piece for the others, ensuring that a highly complicated subject could be explained in easily digestible chunks.
[Copy extract]
Practical applications of reinforcement learning
Autonomous cars – taking the above driving example further, reinforcement learning allows for highly complex and varying scenarios. For example, the weighing up of safety, comfort, the law, emissions, time and speed. It’s simply not possible to predict and thus build-in every single possible scenario and how it might unfold in real life. However, by giving different rewards and penalties, the system can learn how to reach desired goals in circumstances which cannot be totally foreseen. As well as cars, most autonomous lorries, ships and drones are being developed with the aid of reinforcement learning.
Prosthetics – a “Learning to run” project trained a virtual runner – an advanced musculoskeletal model – as the first step in developing a new generation of prosthetic legs. People differ in their walking patterns by often minute amounts, and reinforcement learning can be used to make movement more effective and natural in these circumstances.
Game playing – one of the first and best-known applications of reinforcement learning was when the machine learning algorithm AlphaGo played Go with one of the world’s best human players and won. Reinforcement learning is now used in all kinds of games, both in their development and actual playing.
Medicine – reinforcement learning has many applications in healthcare, from clinical trials through to creating optimal treatments and drug combinations for health conditions. That includes the discovery and generation of dynamic treatment regimes (DTRs) for chronic diseases. It can also be used to quantifying the effects of delaying treatment.
Industry – there are numerous industrial applications for reinforcement learning. These include building intelligence into energy, manufacturing, automotive and supply chains, as well as the optimisation of predictive maintenance. One well-known HVAC example is enabling Google to optimise cooling requirements in their data centres – this has enabled them to reduce energy spending by 40%.
Financial trading – reinforcement learning models can be employed to determine whether to buy, sell or hold stocks based on their market price and other factors. Previously, analysts had to make decisions themselves – not easy given the multiplicity of stocks and the constant fluctuation of market prices.
Online recommendations – while unsupervised learning is good for categorising and prioritising news articles, reinforcement learning goes a step further by tracking reader behaviour to make more accurate recommendations. A wider application is where targeted advertising is delivered to individuals based on their previous purchases or browsing habits.
Leisure – the artificial intelligence research laboratory OpenAI was set up in 2015 to develop “friendly AI” that benefits society. Its research focuses on reinforcement learning, with applications including motor skills, music and gaming. It also supports individuals in creative activities such as writing and composing.
How my copywriting has made a big difference to results and profitability
